Vous êtes sur la page 1sur 16

Chemical Engineering Science 55 (2000) 4101}4116

Overall solids movement and solids residence


time distribution in a CFB-riser
Kathleen Smolders, Jan Baeyens*
Katholieke Universiteit Leuven, Department of Chemical Engineering, de Croylaan 46, 3001 Heverlee, Belgium

Abstract

The residence time distribution (RTD) of solids is essential for the design of CFB reactors where conversion proceeds with time. The
residence time distribution for the solids is measured at di!erent working conditions. The resulting average residence times are
correlated as function of gas velocity and solids circulation rate and are compared with literature data. In order to predict the RTD of
the solids, the solids/gas #ow is "rstly described by a plug #ow with dispersion. A "tting procedure gave experimental PeH clet numbers,
which were correlated as function of gas velocity and solids circulation rate. A more fundamental approach based on the core/annulus
#ow structure is thereafter used to predict the residence time. The riser is divided in a dilute core with particles #owing upward and
a denser "lm moving slowly downward along the wall. Particle interchange between the two regions is described by a convective
interchange #ow. The model is used to predict experimental RTD-curves. The interchange #ux between core and annulus corresponds
well with radial #uxes reported in literature. Although the RTD is qualitatively well described, experimental curves show a higher
dispersion than the calculated ones. To improve the core/annulus approach, further research is necessary.  2000 Elsevier Science
Ltd. All rights reserved.

Keywords: Fluidization; Riser; Solids mixing; Residence time; Core/annulus model

1. Introduction Yerushalmi and Avidan (1985) made RTD-measure-


ments using a ferromagnetic tracer and an inductance
To characterise and model the particle and/or gas #ow bridge detector. Helmrich, Schurgerl and Janssen (1986)
in a riser is not easy although the residence time distribu- reported a double-peak RTD in their experiments using
tion is essential for the design of gas/solid reactions where a radioactive tracer. Bader, Findlay and Knowlton (1988)
solid and/or gas conversions proceed with time and are and Rhodes, Zhou, Hirama and Cheng (1991) used salt as
expected to be as high as possible within the time spent in tracer to be detected by dissolving it and by measuring
the riser. In this paper, measured residence times for the the electrical conductivity of the resulting solution. Their
solids in the CFB are reported and the overall #ow results indicate a high degree of re#ux near the wall and
pattern will "rstly be described by a plug #ow regime substantial solids transfer between riser core and wall
with dispersion. A mechanistic approach based on the regions.
core/annulus #ow structure is thereafter used to predict A similar technique was adopted in this work to deter-
the residence times. mine the residence time in the riser. The experiments
where performed in a riser of 0.1 m ID and 6.5 m height.
The experimental layout and procedures are described
2. Experimental elsewhere (Smolders, 1999; Smolders & Baeyens, 1998).
Essential features are illustrated in Figs. 1 and 2. As
2.1. Experimental procedure illustrated in Fig. 1, the recycle loop entered the side of
the riser at 0.3 m above the distributor plate; constructed
Several studies of solids mixing and residence time as two perforated plates with a "lter paper sandwiched in
have already been published. Avidan (1980) and between and yielding a pressure drop of 10 mbar at
a super"cial velocity of 1 m/s. The tracer injection was
* Corresponding author. Tel.: 32-16-322676; fax: 32-16-322991. located at the height of the recycle entry and penetrated
E-mail address: jan.baeyens@cit.kuleuven.ac.be (J. Baeyens). 5 cm into the riser.

0009-2509/00/$ - see front matter  2000 Elsevier Science Ltd. All rights reserved.
PII: S 0 0 0 9 - 2 5 0 9 ( 0 0 ) 0 0 0 8 4 - 1
4102 K. Smolders, J. Baeyens / Chemical Engineering Science 55 (2000) 4101}4116

The powder was sand of 90 lm. NaCl-salt was used as


tracer. It was milled and sieved to a fraction with
d
150 lm, o
2200 kg/m and a (calculated) trans-
 N
port velocity of 3 m/s, corresponding with the transport
velocity of the sand. As the salt became sticky in a humid
environment, it was stored at 1003C but cooled before
injection. Although transport velocities of sand and salt
were identical, it should be recognised that both powders
might still behave in slightly di!erent way due to di!erent
behaviour during collision and have a di!erent degree of
acceleration.
Prior to starting the RTD-measurements, the gas velo-
city, solids circulation rate and pressure drop pro"le were
measured. The injection system was "lled with 25 g of
salt. The holder is then pressurised. At time zero, both the
Fig. 1. Injection system for the salt tracer. injection valve and the sampling valve are opened simul-
taneously. The tracer is sampled with the sand until the
tube has been "lled. The sampling rate is determined as
the mass of powder in the sampling tube divided by the
sampling time. After sampling, the tube is progressively
emptied in several small cups. Each mixed sand and salt
sample is then dissolved in 100 ml water of which the
electrical conductivity is measured. The electrical con-
ductivity of the solution is transformed into a salt
concentration using a calibration curve. As previously
determined during the validation of the technique, the
presence of the sand did not in#uence the measured
conductivity and hence "ltration is not needed.
The time spent in the exit duct and sample collecting
tube was ;0.1 s and accounted for. It is however negli-
gible when compared to the average time spent in the
riser itself.
Both essential parameters, i.e. the gas velocity and
solids circulation rate, were varied during the tests and
examined combinations of both together with experi-
Fig. 2. Sampling probe for the determination of the solids RTD.
mental results (see Sections 2 and 3 below) are given in
Table 1. Experiments were repeated 3 times. The injec-
tion time of the salt has been measured and was 1 s for
25 g of salt, whereas the total duration of the test was
The exit of the riser was executed as a 903 sharp between 10 and 60 s. Only a fraction of the total solid
connection (Fig. 2). Other inlet and outlet con"gurations #ow is sampled at the riser outlet and only part of the
will a!ect the #ow in the riser, and are currently being injected salt is thus recovered. Since the salt is uniformly
examined. Literature data illustrate this e!ect and are dispersed in the riser outlet, the total sampled and cap-
included in Section 3.3. Continuous sampling at the riser tured mass of salt M is equal to
   
outlet was performed by the probe of Fig. 2. The sample
was collected in the 1.5 cm ID sampling tube. Since the MH
M "M   . (1)
tube was progressively "lled as time proceeded, time-       MH
 
dependent salt concentrations were measured by section-
ing the sample height. For all experimental runs, the Repeatability and consistency of the procedure were
sampling ori"ce was located in the axis of the exhaust proven since calculated (Eq. (1)) and measured salt
pipe (Fig. 2), at 30 cm from the riser wall. It was pre- quantities were within 10%. The same degree of consist-
viously demonstrated that moving the position of the ency was obtained when moving the collection ori"ce
sampling ori"ce across the exit pipe had little e!ect on across the exit duct.
the collected sand}salt powder, thus stressing that there Fig. 3(a) illustrates the measured concentration pro-
was su$cient transverse mixing prior to the point of "les. The tracer is already detected shortly after its injec-
sampling. tion ((1 s) as part of the upward particles #ow in the
K. Smolders, J. Baeyens / Chemical Engineering Science 55 (2000) 4101}4116 4103

Table 1
Results for the RTD-experiments at di!erent operating conditions

; (m/s) G (kg/m s) tM (s) t (s) t (s) (m) t (s) Pe (dimensionless) Pe (dimensionless) S (dimensionless)
      Q  
2.82 5.19 7.75 5.89 4.50 2.9 2.33 5.72 14.7 2.5
2.93 8.72 9.06 6.96 6.02 3.1 2.25 3.88 13.1 2.6
3.04 12.6 11.2 8.46 5.56 3.2 1.96 2.82 11.9 2.5
3.44 4.03 5.37 4.16 2.21 2.3 1.99 9.09 19.3 1.7
3.48 9.74 5.72 4.63 4.26 2.6 2.05 4.51 11.2 2.4
3.45 11.2 5.51 4.77 3.99 2.7 1.84 10.3 26.2 1.6
3.63 15.4 6.77 4.72 3.95 2.7 1.90 2.42 6.41 3.3
3.95 8.67 4.83 3.54 1.96 2.2 1.93 3.79 9.35 2.5
3.79 15.0 4.55 3.57 4.43 2.6 1.97 3.07 6.42 2.5
3.91 23.5 5.81 4.00 3.53 2.7 1.49 2.78 6.72 3.2
4.12 9.55 7.43 4.58 2.23 2.1 2.04 2.75 8.90 3.4
4.37 10.0 3.26 2.54 2.50 2.0 1.95 7.31 12.5 2.0
4.51 14.6 4.20 2.64 2.57 2.1 1.85 3.17 5.83 3.8
4.39 32.6 4.30 3.11 3.87 2.6 1.65 2.68 5.66 3.0
4.85 22.8 5.26 3.31 3.41 2.1 1.96 3.54 10.1 3.4
4.89 17.0 4.17 3.43 2.24 1.9 1.68 3.80 11.3 2.3
4.92 34.1 4.67 3.25 3.18 2.3 1.63 3.67 5.59 2.7

Fig. 4. Convolution principle of block-formed injection and riser sys-


tem.

systems (Levenspiel, 1972)


R
E(t)" E (tH)E (t!tH) dtH. (2)
 
Fig. 3. (a) Measured concentration pro"le (;"4.92 m/s, 
G"34.1 kg/m s; and (b) Resulting residence time distribution.
Furthermore, the overall average residence time of the
total system is equal to the sum of the average residence
time of both parts.
core without transfer to the annulus zone. They are The residence time distribution of the block function is
conveyed at a velocity equal to the gas velocity in the given by


core, which exceeds the super"cial gas velocity. This core 1
#ow of tracer is detected as a "rst peak whereas the , t(2tM ,
E (t)" 2tM  
re#uxing tracer is thereafter detected as a tailing signal.  (3)
Since the peak is already seen within 5 s, the injection 0, t*2t

time period cannot be neglected. The injection is
modelled as a block-function in the data treatment. with tM equal to half the injection time.

The residence time distribution of the total system can
2.2. Data treatment be calculated from the experimentally measured salt con-
centration C as
The injection of the salt is considered to be a tracer
pulse which is dispersed "rstly by an injection block- C(t) C(t )
E(t)" + G . (4)
function and secondly by the riser system itself (Fig. 4). It C(t) dt  C(t )*t
can be proven that the residence time distribution (RTD)  G G G
of the total system is the convolution of the RTDs of both Illustrations of RTD-results are given in Fig. 3(b).
4104 K. Smolders, J. Baeyens / Chemical Engineering Science 55 (2000) 4101}4116

The average residence time is given by reported in Table 1 are based on the values for the total
system of injection and riser.


 
tM " tE(t) dt+ t E(t )*t . (5)
G G G
 G 2.3. Results
Hence, the average residence time for the riser is equal to
When the riser #ow is a pure (dilute) plug #ow, the
t theoretical residence time can be calculated as
tM "tM "tM !tM "tM ! 
. (6)
    2

t " with ; "; . (12)
The results are given in Table 1.  ;/e!;   R
The cumulative residence time distribution F(t) is  
given by For the sand with a terminal velocity of 0.48 m/s, resi-
dence times are predicted to be 3 s or less. As experi-
RG C(t) dt G C(t ) *t mentally determined values lie between 4 and 7 s, the
F(t )"  + H H H , (7)
G C(t) dt  C(t ) *t existence of pure plug #ow in the riser can be ruled out.
 H H H
The values of E(t) at large times have an important
where F(t) is a monotone increasing function. F(t) reaches in#uence on the average residence time tM , whereas
a value of 0.5 at t meaning that 50% of the particle  
  they are of less importance for the value of t . Average
 
#ow has a residence time smaller than t . As can be
  residence times will hence exceed the t -values. Average
 
numerically demonstrated, the value of t for the total
  residence times t of Table 1 are calculated using Eq. (8)
Q
system is equal to the sum of the values of the two over both the total riser and the hydrodynamic fully
separate parts when the riser is modelled by a plug #ow developed part of the riser. Especially the last ones are
with axial dispersion. As a good approach, t for the
  smaller than those obtained from the RTD-experiments,
riser is found by subtracting half the injection time of the which include the bend at the riser outlet. At high gas
t -value of the total system. Results are also given in
  velocities, t approaches the real residence time tM
Q  
.
Table 1. Fig. 5 illustrates the in#uence of the gas velocity and
When assuming that the riser has no dead or bypass solids circulation rate, both on tM and t . As expected,
zones, the average residence time can also be calcu-    
an increasing gas velocity increases the solids velocity
lated from the volume divided by the volumetric #ow of and hence decreases the residence time. The back-mixing
particles of solids is enhanced by increased solids circulation rate,
A; *P which results in an increase of the residence times at
t" or t " 0 (8) equal gas #ow rate. A "tting procedure correlated the
Q A;G/o (1!e ) Q Gg
N average residence time tM as
 
since
G 
tM "56 for a riser height of 6.5 m (13)
*P   ; 
(1!e )" 0 (9)
go
N and t as:
 
The acceleration pressure drop is neglected in Eq. (9), and
correct calculations can hence only be made over the G 
t "60 for a riser height of 6.5 m. (14)
hydrodynamic fully developed region. The length of the   ;
acceleration zone can be calculated according to Wong,
The correspondence between predicted and experimental
Pugsley and Berruti (1992) by
residence times (both own and literature data) is con-

  
D \  o \  "rmed by Fig. 9, as given in Section 3.3.
 "7.92;10 N
D d o
N E

 
G   3. Modelling the riser by plug 6ow with dispersion
; 1# Re\ . (10)
o ; N (Rhodes et al., 1991)
E
The spread in residence time can be expressed as
3.1. The plug yow model with dispersion
t !t
S"     (11) The motion of solids in vertical gas/particle #ow is
t
  very complex. Generally, the #ow pattern consists of
with t and t the times for F(t), respectively, equal to a dilute, rapidly rising core surrounded by a denser,
   
0.1 or 0.9. As the convolution signal is measured, t and slowly descending region near the riser wall. In analogy
 
t are di$cult to assess for the riser only. Results with previous mixing studies in CFB (Rhodes et al., 1991;
 
K. Smolders, J. Baeyens / Chemical Engineering Science 55 (2000) 4101}4116 4105

With this assumption and with the solids PeH clet number
Pe de"ned as
Q
<
Pe " Q  
Q D
X
and with the dimensional time h de"ned as h"t/t ,
 
Eq. (15) becomes

 
C (t (!< t)
C"     exp ! Q  
(t 4D t
X

 
C Pe (1!h)
"   exp ! Q (17)
(h 4h
where C is the tracer concentration at the centre of
 
gravity of the cloud at t"t :
 
C
C "  . (18)
  2(pD t
X  
This PeH clet number Pe is related to the PeH clet number
Q
Pe based on the super"cial gas velocity by
<
Pe "Pe Q   . (19)
Q ;
The RTD of the block function is given by Eq. (3). The
combination of Eqs. (4) and (17) results in the RTD of the
riser. The RTD of the total system is the convolution of
the two as given by Eq. (2).

Fig. 5. In#uence of gas velocity and solids circulation rate on (a) 3.2. Results
average residence time: tM ; and (b) residence time at F(t)"0.5: t .
PGQCP  
The convolution principle was used in a "tting proced-
ure, with the Pe-number as "tting parameter and using
the experimental t value. The Pe numbers were chosen
 
Diguet, 1996) and despite the fact that the process of axial to minimise the least-squares deviations of predicted
solids mixing in such a system is quite di!erent from and experimental RTD-values. Results are illustrated in
molecular di!usion, we investigated the possibility of previous Table 1 and Fig. 6.
describing the mixing phenomena using a one-dimen- The resulting Pe numbers are rather small and vary
sional dispersion model under appropriate assumptions. from 1 to 20. Fig. 7 illustrates how Pe numbers decrease
For one-dimensional di!usion from a point source in with increasing loadings and con"rms the higher back-
a #uid at rest, the tracer concentration at time t and at mixing at higher solid #uxes. The Pe number can be
axial position z is given by correlated by
18.3

 
C z Pe"2.4# (20)
C"  exp ! , (15)  
B
2(pD t 4D t
X X with "G/;, the solids loading (kg/m).
B
Because of the intense transfer between core and
where C is the integration constant and D is the axial
  annulus zone, the high dispersion was expected. As the
dispersion coe$cient. This solution is adapted to the case theory of plug #ow with axial dispersion is only valid for
under investigation by assuming the centre point from Pe-numbers larger than 100, the use of this approach to
which di!usion occurs to move at a velocity equal to the model the particle #ow in the riser is questionable al-
velocity of the centre of gravity of the injected cloud of though it can be used to predict a residence time distribu-
tracer, < or
Q   tion curve that approaches the experiments very well.
Fig. 8 illustrates, e.g. the good agreement between pre-
dicted and experimental t and t values. Using
< "
Q   t
. (16)    
Eq. (17) with the correlations for t (Eq. (14)) and the
   
4106 K. Smolders, J. Baeyens / Chemical Engineering Science 55 (2000) 4101}4116

Fig. 7. The PeH clet number for the dispersion model.

Fig. 6. Illustration of the dispersion model.


Fig. 8. Experimental values for t and t vs. values predicted by the
   
axial dispersion model.

Pe-number (Eq. (20)) gives a good prediction of the solids


residence time distribution in the riser. Once the solids Table 2
RTD is known, the average residence time tM and the Residence times by Diguet (1996)
 
spread in residence time can be calculated. ; (m/s) G (kg/m s) t (s) tM (s) Pe (dimensionless)
   
3.3. Comparison with literature data 4.87 10 4.6 4.6 75
6.20 21 2.4 2.6 35
Literature data on RTD in CFBs are scarce. Similar 6.20 47 2.8 2.9 23
6.18 76 4.1 4.1 25
RTD-experiments were done by Diguet (1996) in a riser 8.14 28 1.7 1.8 40
of 152 mm ID and 6 m height. His results are given in 8.06 42 1.8 1.9 36
Table 2. Comparison with our data is given in Fig. 9. 8.15 78 1.5 1.7 27
Although the gas velocities used by Diguet (1996) are 7.86 117 4.7 3.0 61
higher, residence times correspond very well with our
own and can be predicted by Eqs. (13) and (14). The
Pe-numbers as presented by Diguet (1996) are higher
than predicted by Eq. (20). Whereas the CFB in Leuven Rhodes et al. (1991) investigated the in#uence of the
has a T-shaped end, the riser of Diguet (1996) is provided riser diameter on the dispersion of the solids. Experi-
with a smooth exit. Back-mixing resulting from the ments were done in two CFBs with an inner diameter of,
down#ow of particles will therefore be less than in our respectively, 152 and 305 mm. He concluded that the
experiments. dispersion decreases with increasing riser diameter. The
K. Smolders, J. Baeyens / Chemical Engineering Science 55 (2000) 4101}4116 4107

Fig. 9. Comparison of Eqs. (13) and (14) with own and literature data.

PeH clet-number was correlated as used in the next paragraph where the dispersion will be
predicted by a model based on the core/annulus #ow in
Pe"9.2D(GD) . (21)
the riser.
This correlation of Rhodes overestimates the dispersion
for all experimental cases. Furthermore, the dependence
of Pe with the solids circulation rate is also unexpected as 4. Model based on the core/annulus 6ow
Eq. (21) predicts an increase of the Pe-number with
increasing solids circulation rate whereas experimental 4.1. Introduction
results and Eq. (20) predict the opposite: experimental
observations showed that the down#ow along the riser In the fast #uidization regime, a dilute region with
wall increases with increasing solids circulation rate and rapidly rising particles exists in the core of the riser. This
hence results in a higher back-mixing of particles. These core is surrounded by a denser annulus of slowly de-
observations support Eq. (20) rather than Eq. (21). Part scending particles. A mathematical model, based on the
of the discrepancy between literature results and own observed core/annulus #ow structure, is proposed to
data may be caused by the di!erent riser layout and characterise the #ow and to predict the residence time
particles used. distribution of solids in the riser.
Rhodes et al. (1991) also compared solids RTDs mea- The "rst step of the riser-model development consists
sured in the riser core with those measured at the riser in the characterisation of the two-phase #ow, based on
wall. As the residence times at the wall exceed those in the the principles of the Berruti and Kalogerakis (1989)
core, the core/annulus #ow is con"rmed. This will be model: the riser comprises a dilute core with particles
4108 K. Smolders, J. Baeyens / Chemical Engineering Science 55 (2000) 4101}4116

#owing upward and a denser "lm moving slowly down- The average density along the riser can be obtained from
wards at the wall. Fluxes, voidages and velocities are the measured pressure drop pro"le. The core/annulus
taken constant for each phase at a particular cross- #ow for the steady-state situation can then be numer-
section but can vary along the riser height. In a second ically characterised by solving Eqs. (22)}(27) with known
step, mass balances for the tracer in each phase are parameters G, ;, ; , e and e . Modi"cations are re-
U U 
formulated and solved together with the initial and quired for the unsteady state.
boundary conditions. A sensitivity study thereafter inves- Tracer balances (Ambler, Milne, Berruti and Scott,
tigates the correctness of the model and evaluates the 1990).
in#uence of the di!erent parameters. The calculated mass The injection of a tracer pulse can be described by the
fractions of tracer at the outlet of the riser are compared unsteady state tracer balances for the core and the annu-
with the experimental ones. lar zone. Plug #ow without axial dispersion is assumed
Basic assumptions to characterise the two-phase yow for both zones of the riser (core and annulus). Back-
(Fig. 10). mixing is caused by the exchange of particles between the
The basic assumptions include core and annulus, which is represented by two inter-
change coe$cients I and I .
E particles #ow in the dilute core with a slip velocity ?A A?
A tracer balance around a di!erential volume
equal to their terminal velocity,
(Fig. 11a) in the core zone yields
E the core diameter can vary with riser height,
E voidage and velocities in the core are constant at one
 
* E *f
(E f )#+I f !I f ,"! A A . (28)
height but vary with height, *z A A A? A ?A ? ; *t
E gas travels only in the core, not in the annulus, and QA
E as little is known about the annulus #ow, voidage and The tracer balance around a di!erential volume
particle velocity in the annulus are taken constant over (Fig. 11b) in the annulus zone results in
the whole riser.

 
* = *f
(=f )#+I f !I f ," ? . (29)
This yields a system of the following six equations: *z ? A? A ?A ? ; *t
U
1. G"E !=. (22)
A The interchange coe$cients are not independent since
All #uxes are de"ned as super"cial #uxes a mass balance around a di!erential volume along the
riser gives the net particle #ow rate from the core to the


r 
2. E "; o (1!e ) A , (23)
A QA N A R
3. ; "; !; "; , (24)
  A
EA QA R


; R 
4. ; " , (25)
EA e r
A A

  
r 
5. =" 1! A o (1!e ); , (26)
R N U U


r  e !e
6. A "  U . (27)
R e !e
A U

Fig. 10. Characterisation of the core/annulus #ow. Fig. 11. Tracer balance around the core (a) and the annulus (b).
K. Smolders, J. Baeyens / Chemical Engineering Science 55 (2000) 4101}4116 4109

annulus at any level be chosen on a physical basis or from literature data,


dE whereas the output has to be checked on its physical
I !I "! A . (30) correctness.
A? ?A dz In order to solve the di!erential equations numerically,
Initially, no tracer is found in the riser or all parameters and especially the voidage have to be
f (z, 0)"f (z, 0)"0. (31) known at any height in the riser. The experimental pres-
? A sure drop curves are therefore "tted by a polynomial,
At the bottom of the riser, the newly injected tracer and which is afterwards used to calculate the average voidage.
tracer coming from the annulus will be mixed before Particle velocity and voidage in the annulus layer will
entering the core. A mass balance at z"0 yields in#uence the core/annulus #ow. Literature data on both
mH (t)#f (0, t)="f (0, t)E (32) parameters are however scarce. Farrell, Glicksman and

? A A Hyre (1994) reported a particle velocity of 1 m/s to calcu-
with late the heat transfer at the wall. In accordance with the


m

for t)t , advice of Berruti and Kalogerakis (1989), the terminal


mH " t 
velocity of the sand (+0.5 m/s) is used as a base value in

(33) the calculations. For the solids concentration in the wall
0 for t't .

zone, Berruti and Kalogerakis (1989) recommended the
At the riser outlet, part of the core #ow is re#ected to voidage at minimum #uidisation. The investigations with
form the annulus phase at the top. The weight fractions the ECT-system (Smolders, 1999) however showed the
in both phases are the same at that point solids concentration to be a lot lower. The concentration
can however not be below the average concentration in
f (, t)"f (, t). (34)
? A the riser. A value of 0.1 is used as a base value in the
The system of two conservative hyperbolic partial di!er- calculations. Since the base values are below experi-
ential equations has to be solved together with one initial mental values, the e!ect of the particle velocity and
(Eq. (31)) and two operational boundary conditions voidage are discussed in Section 4.4. An iterative method
(Eqs. (32) and (34)). The mass fraction at the top of the was applied to solve the system of six equations
riser is also the fraction of the tracer in the outlet #ow of (Eqs. (22)}(27)) with the chosen parameters. The cal-
the riser. culated #uxes and the thickness of the annulus are illus-
trated in Table 3 while using values of ; "0.5 m/s and
U
4.2. Practical implementation and preliminary calculations (1!e )"0.1. The small calculated thickness of the an-
U
for a theoretical case nulus ((1 mm) should be interpreted as an incomplete
covering of the wall by the annulus. A linear pressure
To predict experimental results by the model requires drop pro"le was used in this illustration as it results in
some attention. Several variables and parameters have to a constant solids concentration over the riser and hence

Table 3
Results of core/annulus approach

; G d E = Best "t interchange parameter Peak height of calculations


 
(m/s) (kg/m s) (mm) (kg/m s) (kg/m s) (kg/m) for I (kg/m s) divided by peak height of
?A
experiments

2.82 5.19 0.123 5.83 0.64 1.84 1 2


2.93 8.72 0.372 10.66 1.94 2.98 5 2
3.04 12.6 0.442 14.89 2.29 4.14 5 1.7
3.44 4.03 0.055 4.32 0.29 1.17 1 2
3.48 9.74 0.241 10.99 1.25 2.80 5 2
3.45 11.2 0.263 12.57 1.37 3.25 5 1.2
3.79 15.0 0.431 17.24 2.24 3.96 5 1.3
3.63 15.4 0.385 17.40 2.00 4.24 5 2
3.95 8.67 0.081 9.09 0.42 1.27 1 2
4.12 9.55 0.176 10.47 0.92 2.32 5 1.3
3.91 23.5 0.466 25.91 2.41 6.01 10 1.7
4.37 10.0 0.146 10.77 0.77 2.29 5 2.5
4.51 14.6 0.231 15.81 1.21 3.24 5 2
4.39 32.6 0.911 37.30 4.70 7.43 10 1.1
4.89 17.0 0.270 18.41 1.41 3.48 10 2
4.85 22.8 0.558 25.69 2.89 4.70 10 1.1
4.92 34.1 0.758 38.02 3.92 6.93 10 1.1
4110 K. Smolders, J. Baeyens / Chemical Engineering Science 55 (2000) 4101}4116

in constant values for the other parameters For the annulus, calculations with two di!erent values
(E , =, ; ,2). The in#uence of the used pressure drop for the down#ux are done (2 and 10 kg/m s), whereas the
A QA
pro"le will be discussed in Section 4.4. particle velocity in the annulus is always "xed at 0.5 m/s.
Once the characteristics of the riser #ow are known, Interchange parameters have been varied between 0.1
the tracer balances are solved with the package CLAW- and 100 kg/m s. As the interchange parameter from
PACK (Leveque, 1997), which is a collection of Fortran core-to-annulus will generally be the higher of the two
subroutines for solving hyperbolic systems of conserva- (Eq. (30)), calculations with I equal to twice I were
A? ?A
tion laws in one and two dimensions, both with and also performed. Simulation results are presented in
without source terms. The accuracy of the package was Fig. 12. The shape of the simulation curves is mainly
"rst tested with one partial di!erential equation, which determined by the interchange coe$cients. When ex-
describes pure convection #ow without dispersion. The change of particles between core and annulus is high, the
injected block pulse returns at the exit with almost no weight fraction of the tracer will be the same in both
deformation. phases. High interchange coe$cients will therefore pre-
Thereafter, preliminary calculations were made to dict more or less plug #ow behaviour of the solids. At
check the validity of the tracer balances. Characteristic small interchange parameters, the #ux of tracer from core
values of #uxes and velocities are chosen and kept con- to annulus can still be great, as it is proportional to the
stant along the riser. Solid #ux and particle velocity in weight fraction f in the core. The opposite #ux from
A
the core are taken as 40 kg/m s and 4 m/s, respectively. annulus to core however is very small, as f remains
?

Fig. 12. Preliminary calculations of the RTD using the core/annulus approach.
K. Smolders, J. Baeyens / Chemical Engineering Science 55 (2000) 4101}4116 4111

small. Tracer entering the annulus can in this case only


return to the core at the bottom of the riser. This can be
interpreted as a new injection pulse, which causes a sec-
ond peak to appear in the RTD-curves.
The right side of the curve is generally the least steep,
as the tracer is spread and back-mixed by the annulus
layer. The steep left side of the curve corresponds to
the detection of the particles that are not transferred to
the annulus. The time needed to detect a considerable
amount of tracer increases at higher exchange rates as
fewer tracer particles remain in the core.
Fig. 12 clearly shows that increasing the down#ow will
enhance the back-mixing as more particles will com-
pletely return to the bottom of the riser. Simulations with
the values of the core-to-annulus interchange coe$cient
to be twice the annulus-to-core interchange coe$cient
result in wider residence time distributions, as could be
expected. As the upward #ux E in our experiments varies
A
only slightly along the riser height, the two interchange
parameters will always be around the same value. Fur-
ther study with two di!erent interchange parameters is
not needed here. The results observed in Fig. 12 justify
the use of the core/annulus procedure in the prediction of
the residence time distribution.

4.3. Prediction of the experimentally determined


residence time distributions

After the brief parameter study for a theoretical case,


calculations for the experimental results have been per-
formed. As illustrated before, the interchange coe$cients
determine the shape of the curves and will be varied to "t
the experimentally measured curves. The annulus-to- Fig. 13. Prediction of the residence time distribution using the
core/annulus approach; (a) ;"3.48 m/s, G"9.74 kg/m s; (b)
core interchange parameter is assumed to be constant ;"4.85 m/s, G"22.8 kg/m s; (c) ;"4.92 m/s, G"34.1 kg/m s.
along the riser, whereas the core-to-annulus interchange
parameter can be calculated by Eq. (30). Results for
di!erent values of the interchange parameter are given in
Fig. 13. The experimental weight fraction is an average of
the weight fractions, collected during the time interval *t.
Calculated curves are averaged in the same manner. For
all experimental cases, it is obvious that the shape of the
experimental curves can be correctly predicted, although
the numerical values are not always accurate. The cal-
culated curves always show a higher peak and a less
pronounced tail than the experimental ones. This results
in higher t and smaller t values as illustrated in
   
Fig. 14.
When investigating the curves qualitatively, the best "t
interchange parameter seems to increase with increasing
solids circulation rate


1 kg/m s for G(5 kg/m s,
I + 5 kg/m s for 5 kg/m s(G(17 kg/m s,
?A Fig. 14. Experimental values for t , t and t vs. predicted values
     
10 kg/m s for G'17 kg/m s. by the core/annulus approach.
4112 K. Smolders, J. Baeyens / Chemical Engineering Science 55 (2000) 4101}4116

This interchange parameter I can be related to an Di!erences between experimental and calculated
?A
interchange #ux IH based on the interface surface curves can be caused by experimental errors and/or im-
?A
2pr perfections in the model. Firstly, the accuracy of the
I "IH ; A , (35) experimental tests is su$cient to de"ne the average resi-
?A ?A A dence times and the shape of the distribution, but does
with IH as the interchange #ux (kg material/m inter- not provide exact weight fractions. Secondly, the samp-
?A
face s) ling of the salt/sand #ow occurred after the exit and bend
As I varies between 1 and 10 kg/m s, IH will vary of the riser outlet. The change of direction in #ow can
?A ?A
between 0.025 and 0.25 kg/m s. cause extra dispersion of tracer particles, which is not
The value of this interchange #ux corresponds well accounted for in the calculations. This extra back-mixing
with measured radial #uxes found in literature. For a gas can be an important portion of the dispersion, especially
velocity of 5.5 m/s and a solids circulation rate of at low solid loadings where the core/annulus dispersion
40 kg/m s, Zhou et al. (1995) measured radial #uxes is small. Sampling before the riser outlet however is
between 0.1 kg/m s in the core zone and 1.5 kg/m s at di$cult as the composition of the #ow will vary across
the wall (Fig. 15). the riser section. Furthermore, the calculation procedure
For a similar experiment, our model predicts a best "t comprises the choice of parameters without experimental
interchange #ux of 0.25 kg/m s in line with Zhou's data. validation. Changing these parameters without violating
The measured radial #uxes of Zhou et al. (1995) increase physical limits can improve the predictions as will be
with solids circulation rate, which is also observed in the examined in the next section.
predicted values of I .
?A
Looking quantitatively at the predicted RTD-curves, 4.4. Sensitivity analysis
predictions improve when the solid loading is higher at
a speci"c gas velocity. Best "t values of I and accuracy The complexity of the model obscures the in#uence of
?A
of the predictions are illustrated in Table 3. the di!erent parameters on the predicted results. To
investigate the possibility of improving the predictions,
an experimental case was calculated with di!erent para-
meters. The study covers the in#uence of
E height of riser,
E pressure drop pro"le,
E solids concentration in the annulus (1!e ),
U
E particle velocity in the annulus (; ),
U
E down#ux of annulus (=),
E particle velocity in the core (; ).
QA
Increasing the height of the riser in the calculation pro-
cedure is illustrated in Fig. 16(a). The area and possibility
of interchange between core and annulus and back-mix-
ing will increase. The arrival time of the "rst detected
particles will also increase. Increasing the height in the
calculations will however not improve the predictions.
The pressure drop proxle determines the core/annulus
characteristics, e.g. core diameter, down#ux, velocity, etc.
Measured pressures were "tted by a fourth-order poly-
nomial for use in the model. Using a linear pro"le, with
almost the same accuracy for the experimental pressure
drop pro"le, gives slightly di!erent weight fraction
curves. Increasing the pressure gradient and hence the
average solids concentration does not signi"cantly alter
the predicted residence time distributions as illustrated in
Fig. 16(b). Very accurate voidage pro"les are hence not
necessary for this predictive model.
Once the voidage pro"le and injection time and mass
are known, the solution of the tracer balances are exclus-
ively determined by the particle #ux and velocity in both
Fig. 15. Radial solids #uxes measured by Zhou (1995); (a) ;"5.5 m/s; riser sections. These characteristics are completely deter-
(b) G"40 kg/m s mined by the choice of the voidage and particle velocity
K. Smolders, J. Baeyens / Chemical Engineering Science 55 (2000) 4101}4116 4113

Fig. 16. In#uence of model parameters on predictions by the core/annulus approach (;"3.92 m/s, G"34.1 kg/m s, I "10 kg/m s): (a) height of
?A
the riser; (b) pressure drop pro"le; (c) solids concentration in the annulus; (d) particle velocity in the annulus; (e) down#ow in the annulus; and (f)
particle velocity in the core.

in the annulus. Fig. 17 illustrates the in#uence of both the core will thus be smaller. Particles in the annulus
parameters on the particle down#ow. reach the bottom of the riser faster with increasing par-
Increasing the solids concentration from 0.05 to 0.4 has ticle velocity and are thus faster re-injected in the core.
only a minor e!ect on the particles down#ow. Varying Both counteracting e!ects can explain the small vari-
e will therefore not improve the "tting of the experi- ations in the predicted residence time distributions
U
mental curves (Fig. 16(c)). (Fig. 16(d)).
On the other hand, increasing the particle velocity will To check the in#uence of the downyux, calculations
increase the #ux in the annulus and hence in the core. with di!erent values for = were performed, where the
This is illustrated in Fig. 17 where the particle velocity in #ux was kept constant across the riser. Whereas E was
the annulus is varied between 0.5 and 2 m/s. The down- also changed to ful"l Eq. (22) ; and ; were constant.
QA U
#ux variation can be explained by the combination of Increasing only the down#ux will increase the back-
Eqs. (26) and (27), assuming e to be almost 1: mixing as transfer from annulus to core is proportional
A
to the weight fraction in the annulus. Increasing the
=+(e !e )o ; . (36)
A  N U down#ux results in a lower weight fraction and tracer
As the particle velocity appears also directly in the tracer particles will remain in the annulus longer (Fig. 16(e)).
balance, predictive results of the RTD may vary with this A last parameter is the particle velocity in the core.
velocity. Increasing the velocity increases the down#ow, Increasing this parameter decreases the time needed to
which leads to higher back-mixing as the transfer from detect the "rst tracer (Fig. 16(f )). At high interchange
annulus to core is proportional to f . The transfer back to parameters, the back-mixing will be reinforced by smaller
?
4114 K. Smolders, J. Baeyens / Chemical Engineering Science 55 (2000) 4101}4116

Fig. 17. In#uence of particle velocity and voidage in the annulus on the down#ux: ;"4.92 m/s, G"34.1 kg/m s.

velocities, as more time for the core-to-annulus transport


is available. At small interchange parameters, the in#u-
ence of the particle velocity is less pronounced. Remem-
ber that the last two parameters are not input parameters
of our model and cannot as such be altered.
In this section, the in#uence of several parameters was
discussed. They all have a weak or strong e!ect on the
calculated results, which can be physically explained.
Varying the parameters however does not signi"cantly
alter the prediction of the experimental residence time
distributions.

4.5. Prediction of literature data (Diguet, 1996)


Fig. 18. Prediction of the core/annulus model for literature data
Fig. 18 illustrates the core/annulus model approach for (Diguet, 1996): (;"4.78 m/s, G"10.0 kg/m s).
experiments performed by Diguet (1996). As only E(t)-
curves are reported, calculated fraction curves are also
transformed into residence time distribution curves. The of the gas velocity and solids circulation rate. Although
solids were captured before the bend, which can explain the high degree of back-mixing results in rather small
the slightly better predictions of the model. The best "t PeH clet numbers, the model gives a fairly good prediction
value for the interchange parameters I lies between of the RTD and the average residence time. The experi-
?A
5 and 10 kg/m s and this is slightly lower than the best mental results are con"rmed by literature data.
"t value for our experiments at similar conditions. A mechanistic approach based on the core/annulus
Eq. (35) yields an interchange #ux IH between 0.2 and #ow structure is thereafter used to predict residence times
?A
0.4 kg/m s, comparable with our own experiments. on a more fundamental basis. The riser is divided in
a dilute core with particles #owing upward and a denser
"lm moving slowly downward along the wall. Particle
5. Conclusions and recommendations interchange between the two regions is described by
a convective interchange #ow. The solution of the un-
As conversion proceeds with time in CFB reactors, the steady state mass balances for the tracer gives the mass
residence time distribution of the solid phase is very fraction at the outlet of the riser. The model is used to
important. The residence time distribution is measured at predict the experimental values. The interchange #ux
di!erent working conditions. Eqs. (13) and (14) can be between core and annulus corresponds well with radial
used to predict average residence times. #uxes reported in literature.
The RTD can be described by a plug #ow with disper- The core/annulus model approach has demon-
sion model. Eq. (20) gives the PeH clet-number as function strated to be a practical tool for the prediction of the
K. Smolders, J. Baeyens / Chemical Engineering Science 55 (2000) 4101}4116 4115

back-mixing of solids in a riser #ow. Good qualitative MH captured mass #ow of sand and salt, g/s
 
agreement between predicted and calculated values is M mass of captured salt, kg
   
found with an interchange #ux IH around 0.2 kg/m s. m mass of injected salt, g
?A 

To quantitatively improve results, further research is mH salt injection mass #ow g/s


needed and should include: MH mass #ow in the riser g/s


 
M mass of injected salt, g
 
E measuring model parameters as particle velocity (pre- Pe PeH clet number based on ;(";/D ), dimen-
X
ferably introduced as 1}1.5 m/s, which are common sionless
experimental results) and voidage in the annulus to Pe PeH clet number based on < ("< /D ),
Q       X
justify the used values; dimenionless
E replacing the T-shaped riser exit by an exit where the R radius of the riser, m
riser diameter gradually decreases before ending in r radius of core, m
A
a bend. This will o!er the possibility to measure solids Re particle Reynolds number ("o ;d /k), di-
N  
residence time distributions more accurately without mensionless
the extra back-mixing and obstruction in the exit; S spread in residence time (Table 1), dimension-
E measuring radial #uxes will justify the predicted inter- less
change parameters; t time, s
E repeating experiments for a greater range of gas vel- tM average residence time of total system, s
ocities and solid #uxes in risers of di!erent diameters tM average residence time of injection, s

and with particles of di!erent size will also give a better t residence time with F(t )"0.5, s
   
insight in the interchange parameters. t injection time, s
GLH
t residence time calculated from pressure drop
Q
measurements, s
Notation tM average residence time in the riser, s
 
t theoretical residence time, s

A riser surface area, m ; super"cial gas velocity, m/s
C(t) salt concentration, g/g ; absolute gas velocity in the core, m/s
EA
D riser or bed diameter, m ; particle velocity in the core, m/s
QA
d particle size, m ; slip velocity, m/s
N  
D axial dispersion coe$cient, m/s ; slip velocity in the core, m/s
X   A
E upward solid #ow in the core, kg/ms ; terminal velocity of the particle, m/s
A R
E(t) residence time distribution of injection and ; particle velocity in the annulus, ms
U
riser system, dimensionless < velocity of particle with residence time t , m/s
    
E (t) residence time distribution of the block injec- = downward solid #ow in the annulus, kg/m s

tion, dimensionless z height from distributor, m
E (t) residence time distribution of the riser system, *P pressure drop across the riser, Pa
 0
dimensionless e voidage, dimensionless
F(t) cumulative residence time distribution, dimen- e average voidage, dimensionless
sionless e average voidage across the cross-section of the

f mass fraction of tracer in the annulus, riser, dimensionless
?
g salt/g material e voidage in the core, dimensionless
A
f mass fraction of tracer in the core, g salt/g ma- e voidage in the annulus, dimensionless
A U
terial h time, dimensionless
G solids circulation rate, kg/m s o gas density, kg/m
E
g gravitational constant, m/s
I interchange parameter from annulus to core,
?A References
kg/m riser s
IH interchange #ux from annulus to core, kg/m
?A Ambler, P. A., Milne, B. J., Berruti, F., & Scott, D. S. (1990). Residence
interface s time distribution of solids in a circulating #uidized bed: experi-
I interchange parameter from core to annulus,
A? mental and modelling studies. Chemical Engineering Science, 45,
kg/m riser s 2179}2186.
IH interchange #ux from core to annulus, kg/m Avidan, A. A. (1980). Bed expansion and solids mixing in high velocity
A? yuidized beds. Ph.D dissertation, City College of New York.
interface s
Bader, R., Findlay, J., & Knowlton, T. M. (1988). Gas/solids #ow
riser height, m patterns in a 30.5 cm diameter circulating #uidized bed. In P. Basu,
length of acceleration zone, m
 & J. F. Large, Circulating yuidized bed technology II (pp. 123}137).
solids loading, kg/m Oxford: Pergamon Press.
B
4116 K. Smolders, J. Baeyens / Chemical Engineering Science 55 (2000) 4101}4116

Berruti, F., & Kalogerakis, N. (1989). Modelling the internal #ow Rhodes, M. J., Zhou, S., Hirama, T., & Cheng, H. (1991). E!ects of
structure of circulating #uidized beds. Canadian Journal of Chemical operating conditions on longitudinal solids mixing in a circulating
Engineering, 67, 1010}1014. #uidized bed reactor. A.I.Ch.E. Journal, 37, 1450}1458.
Diguet, S. (1996). Contribution a% l 'e& tude de la de& shydratation du carbon- Smolders, K. (1999). Hydrodynamics of circulating yuidized beds. Ph.D.
ate de sodium monohydrate en lit yuidise& dense et mise en oeuvre en lit dissertation, KULeuven, Belgium.
yuidise& circulant: e& tude expe& rimentale et mode& lisation. Ph.D. disserta- Smolders, K., & Baeyens, J. (1998). Hydrodynamic modelling of circula-
tion, L'Institut National Polytechnique de Toulouse, France. ting #uidized beds. Advances in Powder Technology, 9, 17}38.
Farrell, P. A., Glicksman, L. R., & Hyre, M. R. (1994). Dynamic Yerushalmi, J., Avidan, A. A. (1985). High velocity #uidization. In
similarity in #uidization. personal communication. J.F. Davidson, R. Clift, & D. Harrsion, Fluidization (2nd ed.),
Helmrich, H., Schurgerl, K., & Janssen, K. (1986). Decomposition (pp. 226}293). New York: Academic Press.
of NaHCO in laboratory and bench scale circulating #uidized Wong, R., Pugsley, T., & Berruti, F. (1992). Modelling the axial voidage

bed reactors. In P. Basu, Circulating yuidized bed technology pro"le and #ow structure in risers of circulating #uidized beds.
(pp. 161}166). Toronto: Pergamon Press. Chemical Engineering Science, 47, 2301}2306.
Levenspiel, O. (1972). Non-ideal #ow. In O. Levenspiel, Chemical reac- Zhou, J., Grace, J. R., Lim, C. J., Brereton, C. M. H., Qin, S., & Lim, K.
tion engineering (2nd ed.), (pp. 253}325). New York: Wiley. S. (1995). Particle cross-#ow, lateral momentum #ux and lateral
Leveque, R. J. (1997). Wave propagation algorithms for multi-dimen- velocity in a circulating #uidized bed. Canadian Journal of Chemical
sional hyperbolic systems, CLAWPACK. Journal of Computational Engineering, 73, 612}619.
Physics, 131, 327}353.

Vous aimerez peut-être aussi